and the Laplacian on polygons
Luca Guido Molinari1
Physics Department “Aldo Pontremoli”
Universit`a degli Studi di Milano and I.N.F.N. sez. Milano Via Celoria 16, 20133 Milano, Italy
Abstract: Several sums of Neumann series with Bessel and trigono- metric functions are evaluated, as finite sums of trigonometric func- tions. They arise from a generalization of the Neumann expansion of the eigenstates of the Laplacian in regular polygons. A simple accurate approximation of J0(x) is found, on the interval [0, 2].
MSCS [2020]: Primary 33C10, Secondary 35J05.
Keywords: Bessel function, Neumann series, Laplace equation in poly- gon.
Resubmitted 5 apr 2021
Introduction
The ground state of the Laplace equation in a regular polygon with Dirichlet boundary conditions at the n sides, has a natural expression as a Neumann series of Bessel and trigonometric functions,
ψn(r, θ) = J0(λnr) + 2X∞
k=1hk,nJkn(λnr) cos(knθ),
with coefficients hk,n to be found and eigenvalue −λ2n that scales with the area. For the equilateral triangle and the square, the solutions are known as sums of few trigonometric functions of the coordinates x = r cos θ and y = r sin θ. Such solutions have a corresponding Neumann expression [7]. For the square of area π:
J0(r√
2π) + 2X∞
k=1J4k(r√
2π) cos(4kθ)
= 12cos(x√
2π) + 12cos(y√
2π) (1)
1Luca.Molinari@unimi.it
1
J0(λ3r) + 2X∞
k=1
cos(kπ/2 − π/6)
cos(π/6) J3k(λ3r) cos(3kθ)
= 3√23sin(3R4π
3x + 2π3 ) −3√23[sin[3R2π
3(x + y√
3) −2π3 ]
−3√23sin[3R2π
3(x − y√
3) −2π3 ] (2) where, for area π, λ23 = 4π/√
3 and R3 = 23p π√
3. In [7] I obtained a sum that generalizes the integrable cases n = 3, 4:
fn(x, y) =J0(r) + 2X∞
k=1
cos[nk3π2 −2nπ ]
cos(2nπ) Jnk(r) cos(nkθ)
=1 n
Xn−1
`=0
cos[r cos(θ + 2πn`) + 2nπ] cos(2nπ )
=1 n
Xn−1
`=0
cos[x cos(2πn`) − y sin(2πn`) +2nπ ]
cos(2nπ ) (3)
For n → ∞ the Riemann sum in the second line is R2π 0
dt
2πcos(r cos t) = J0(r). For n = 2 it is f2(x, y) = cos x. For n = 6:
f6(x, y) =J0(r) + 2X∞
k=1(−1)kJ6k(r) cos(6kθ)
=13cos x + 23cos(12x) cos(
√ 3
2 y) (4)
The functions fnare eigenfunctions of the Laplace operator with eigen- value −1 but, for n > 4, they no longer vanish on the boundary of a n-polygon. I only remark that the level curves fn(x, y) = C are closed around the origin (where fn= 1) up to a separatrix with n self- intersections, with values C5 = −0.334909, C6 = −1/3, C7 = 0.19633, etc. The level lines and the separatrices for n = 6, 7 are shown in Fig.1.
The study of the Laplacian in polygons has a long history. The ground states beyond the square, n > 4, cannot be finite sums of trigonometric functions. They have been investigated analytically and numerically in 1/n expansion (see for example [7, 4, 3]).
In this paper I generalize the identity (3), and obtain a number of new formulas for Neumann series whose sums contain a finite num- ber of terms. For certain values of the parameters, they are identities that are found in the tables by Gradshteyn and Ryzhik [2], Prudnikov, Brychkov and Marichev [9], a recent paper by Al-Jarrah, Dempsey and Glasser [1], and two old papers by Takizawa and Kobayasi [10, 5]. In the last ones, the Neumann series appear as correlation functions for the heat flow in coupled harmonic oscillators.
A side result is an accurate approximation of J0(x) on the interval [0, 2],
-5 0 5 -5
0 5
-5 0 5
-5 0 5
Figure 1. Left: the contour plot of the func- tion f6 in eq.(4). Right: the separatrix f6(x, y) =
−1/3 is a Kagom´e lattice. It can be written as 0 = cos(x/2)[cos(x/2) + cos(√
3y/2)]. Vertices are the dou- ble zeros.
-5 0 5
-5 0 5
-5 0 5
-5 0 5
Figure 2. Left: the contour plot of f7 in eq.(3). Right:
the separatrix f7(x, y) = −1.9633...
by three cosines (eq.18).
All figures and several checks were made with Wolfram Mathematica7.
The summation formula The source equation of various sums in this paper is:
+∞
X
k=−∞
Jkn+p(z)eikny = 1 n
n−1
X
`=0
eiz sin(y+2πn`)−ip(y+2πn`) (5)
even n eq.(5) is eq.1 in [10]. Sums of this sort are tabulated only for n = 1, 2 in [9].
Proof. The result follows from the Fourier integral of a Bessel function of integer order. For z ∈ C, the sumP∞
k=−∞eiknyJkn+p(z) is uniformly convergent in y by the bound |J±m(z)| ≤ C|z/2|m/m! (Nielsen, see
§3.13 in [11]).
∞
X
k=−∞
eiknyJkn+p(z) =
∞
X
k=−∞
eikny Z 2π
0
dθ
2πeiz sin θ−i(kn+p)θ
=
∞
X
k=−∞
eikny
n−1
X
j=0
Z 2πn(j+1)
2π nj
dθ
2πeiz sin θ−ipθe−iknθ
=
n−1
X
j=0
" ∞ X
k=−∞
eikny Z 2πn
0
dθ
2πeiz sin(θ+2πnj)−ip(θ+2πnj)e−iknθ
#
The functions pn/2π eikny, k ∈ Z, are a complete orthonormal ba- sis in the Hilbert space L2(0, 2π/n). The infinite sum is the Fourier representation of (1/n) exp[iz sin(y + 2πnj) − ip(y + 2πnj)].
In the following I give some examples.
1. The case p = 0 and y = π2 + α is an extension with angle α of the equations 19 and 20 in [1], where α = 0. With J−m(z) = (−)mJm(z):
J0(z) + 2
∞
X
k=1
eiknπ2Jkn(z) cos(knα) = 1 n
n−1
X
`=0
eiz cos(α+2πn`) (6)
For n = 1, separation of even and odd parity terms in z gives the Jacobi expansions (eqs. 5.7.10.4 and 5 in [9]):
J0(z) + 2
∞
X
k=1
(−)kJ2k(z) cos(2kα) = cos(z cos α) (7)
∞
X
k=0
(−)kJ2k+1(z) cos[(2k + 1)α] = 12sin(z cos α) (8)
If α is replaced by α + π/2 they are eqs. 8.514.5 and 6 in [2] and 10.4, 10.5 in [6]:
J0(z) + 2
∞
X
k=1
J2k(z) cos(2kα) = cos(z sin α) (9)
∞
X
k=0
J2k+1(z) sin[(2k + 1)α] = 12sin(z sin α) (10)
1.1. For n replaced by 2n, eq.(6) is:
J0(z) + 2
∞
X
k=1
(−)knJ2kn(z) cos(2knα) = 1 n
n−1
X
`=0
cos[z cos(α + πn`)] (11) where the finite sum was reduced by noting that terms ` and n + ` are the same.
- The value y = π2 yields eq.(23) in [1].
- For n = 1 the derivative of (11) in α = π4 is:
2J2(z) − 6J6(z) + 10J10(z) − 14J14(z) + ... = z
√2 4 sin(z
√2
2 ) (12) - For n = 2 eq.(11) becomes
J0(z) + 2X∞
k=1J4k(z) cos(4kα) = 1
2[cos(z sin α) + cos(z cos α)]. (13) The values α = 0,π4 give eqs.5.7.1.19 in [9]. The derivative is:
∞
X
k=1
kJ4k(z) sin(4kα) = z
16[sin(z sin α) cos α − sin(z cos α) sin α], (14) further, the expansion in small α gives:
∞
X
k=1
k2J4k(z) = z
64(z − sin z) (15)
- Case n = 3 is eq.(4), α = 0 gives eq. 5.7.1.21 in [9].
1.2. If n is replaced by 2n + 1, eq.(6) is:
J0(z) + 2
∞
X
k=1
ei(2n+1)kπ2J(2n+1)k(z) cos[(2n + 1)kα] = 1 2n + 1
2n
X
`=0
eiz cos(α+
2π 2n+1`)
J0(z) + 2
∞
X
k=1
(−)kJ(4n+2)k(z) cos[(4n + 2)kα]
= 1
2n + 1
2n
X
`=0
cos[z cos(α +2n+12π `)] (16)
2
∞
X
k=0
(−)n+kJ(2n+1)(2k+1)(z) cos[(2n + 1)(2k + 1)α]
= 1
2n + 1
2n
X
`=0
sin[z cos(α +2n+12π `)] (17) The first equation with n = 1 and α = π/12, offers a simple approx- imation of J0(x) in 0 ≤ x ≤ 2 with an error || ≤ 4 × 10−9
J0(x) = 13[cos(x√1
2) + cos(x
√ 3−1 2√
2 ) + cos(x
√ 3+1 2√
2 )] + (18) At j0,1 ≈ 2.404 (first zero of J0(x)) the error is 3.4 × 10−8.
Examples of the second equation, (17), are:
∞
X
k=0
(−)kJ6k+3(z) cos[(6k + 3)α] = −1 6
2
X
`=0
sin[z cos(α +2π3 `)] (19)
∞
X
k=0
(−)kJ10k+5(z) cos[(10k + 5)α] = 1 10
4
X
`=0
sin[z cos(α +2π5 `)] (20) The first one with α = π is eq.22 in [1].
Both sums are eigenfunctions of the Laplacian with eigenvalue λ = −1 (see Figs. 3,4). The finite sum in (20), with z = r, x = r cos α and y = r sin α, is:
f (x, y) = 101 sin x − 15sin(x cosπ5) cos(y sinπ5)
+15sin(x cos2π5 ) cos(y sin2π5 ) (21) 1.3. Eq. (5) with p = 0 is multiplied by exp(iβ), β real, and the real part is taken. The left hand side becomes:
J0(x) cos β +X∞
k=1Jkn(x) Re[eiβ(eikny + e−ikn(y+π))]
= J0(x) cos β + 2X∞
k=1cos(β − knπ2)Jkn(x) cos[kn(y + π2)]
The identity (3) is obtained, when β = 2nπ and y + π2 = θ + π.
-5 0 5 -5
0 5
-5 0 5
-5 0 5
Figure 3. Contour plot of the sum (19). The function is the ground state of the Laplacian, vanishing on the boundary of equilateral triangles (no nodal lines) and is an excited state of the hexagon (the right figure is the contour for the value zero).
-5 0 5
-5 0 5
-5 0 5
-5 0 5
Figure 4. Left: contour plot of the sum (20), i.e the function f in (21). Right: the contour f = 0 is very close to the circumference of radius j5,1 (the first zero of J5), where |f (x, y)| ≤ 0.001.
2. Parseval’s identity is applied to (5):
X
k∈Z
Jkn+p2 (x) = 1 n2
X
k`
eip2πn(k−`) Z 2π
0
dy
2πeix sin(y−πn(k−`))−ix sin(y+π n(k−`))
= 1 n2
X
k`
eip2πn(k−`) Z 2π
0
dy
2πe−i2x cos y sin(π n(k−`))
= 1 n2
X
k`
eip2πn(k−`)J0[2x sin(πn(k − `))]
= 1 n + 2
n2
n−1
X
k=1
k cos(2πnkp)J0(2x sinπkn)
X
k∈Z
Jkn+p2 (x) = 1 n + 1
n
n−1
X
k=1
cos(2πnkp)J0(2x sinπkn) (22) The left-hand side is Jp(x)2+P∞
k=1[Jkn+p2 (x) + Jkn−p2 (z)].
2.1. If n → 2n and p = 0, the sum (22) is amenable to eq.29 in [1]:
J02(x) + 2
∞
X
k=1
J2kn2 (x) = 1 2n + 1
2nJ0(2x) + 1 n
n−1
X
k=1
J0(2x cos2nπk) (23) - If n → 2n and p = n in (22), with simple steps one obtains:
∞
X
k=0
J(2k+1)n2 (x) = 1
4n + (−)n
4n J0(2x) + 1 2n
n−1
X
`=1
(−1)`J0(2x sin2nπ`) (24) Examples:
X∞
k=0J2+4k2 (x) = 18 +81J0(2x) − 14J0(x√
2) (25)
X∞
k=0J3+6k2 (x) = 121 − 16J0(x) −121J0(2x) + 16J0(x√
3) (26) 3. In eq.(5) the variable y is shifted to y+2t. The equation is multiplied by eiz0sin y−iqy and integrated in y:
+∞
X
k=−∞
Jp+kn(z)Jq−kn(z0)ei(kn+p)2t
=1 n
n−1
X
`=0
e−ip2πn` Z 2π
0
dy
2πeiz sin(y+2t+2π
n`)+iz0sin y−i(p+q)y
(27) In the integral, the shift y to y − t − πn` changes the exponent to i(z + z0) sin y cos(t +πn`) + i(z − z0) cos y sin(t +πn`) − i(p + q)(y − t −πn`) 3.1. With z = z0 we obtain eq.1 in [5]:
+∞
X
k=−∞
Jp+kn(z)Jq−kn(z)e2iknt = 1 n
n−1
X
`=0
e−i(p−q)(t+πn`)Jp+q[2z cos(t +πn`)]
(28) For n = 1, with a shift of the index k and renaming of parameter, it is eq.8.530 in [2].
For n = 2, p = q and t = π/4 it is eq. 5.7.11.25 [9].
3.2. Eq.(27) with p = q = 0 and t = 0 is:
J0(z)J0(z0) + 2
∞
X
k=1
(−)knJkn(z)Jkn(z0) = 1 n
n−1
X
`=0
Z 2π 0
dy
2πeiz sin(y+2πn`)+iz0sin y (29) For n = 1, 2 they are eqs. 5.7.11.1 and 5.7.11.18 in [9]; if also z = z0 they are eqs. 31, 32 in [1].
A new example is:
∞
X
k=1
J4k(x)J4k(y) = 18[J0(x + y) + J0(x − y) − 4J0(x)J0(y)]
+14J0(p
x2+ y2) (30) 4. Multiplication of (5) by exp(−ay) (a > 0) with p = 0 and n = 1, and integration on R+ give:
∞
X
k=−∞
J2k(z) a
a2+ 4k2 + 2i
∞
X
k=0
J2k+1(z) 2k + 1 a2+ (2k + 1)2 =
Z ∞ 0
dy eiz sin y−ay
The last integral is done by expanding exp(iz sin y) and using integrals eqs.3.895.1 and 3.895.4 [2]. The even and odd terms are:
1
a2J0(z) + 2
∞
X
k=1
J2k(z) a2+ 4k2 =
∞
X
k=0
(−)kz2k
a2(a2+ 4)...(a2+ 4k2) (31)
∞
X
k=0
J2k+1(z) 2(2k + 1) a2+ (2k + 1)2 =
∞
X
k=0
(−)kz2k+1
(a2+ 1)(a2+ 9)...(a2+ (2k + 1)2) (32) For a = iν the equations are expansions of the Lommel functions s−1,ν(z) and s0,ν(z) in series of Bessel functions (eq. 11.9.7 in [8]).
More and more identities can be obtained by derivation, or integra- tion with functions. Here I limited myself to some examples.
Data availability. Data sharing is not applicable to this article as no new data were created or analyzed in this study.
References
[1] A. Al-Jarrah, K. M. Dempsey, M. L. Glasser, Generalized series of Bessel functions, J. Comp. Appl. Math. 143 (2002) 1–8.
[2] I. S. Gradshteyn and I. M. Ryzhik, Table of integrals, series and products, 7th edition, Academic Press
Dirichlet boundary conditions, arXiv:1712.06082 [math.NA]
R. S. Jones, Computing ultra-precise eigenvalues of the Laplacian within poly- gons, Adv. Comput. Math. 43 1325–1354 (2017).
[4] P. Grinfeld and G. Strang, Laplace eigenvalues on regular polygons: a series in 1/N, J. Math. Anal. Appl. 385 (2012) 135–149.
[5] K. Kobayasi and ´Ei I. Takizawa, On an infinite sum of two Bessel functions of first kind, Chinese Journal of Physics 3 n.1 (1965) 69–71.
[6] B. G. Korenev, Bessel functions and their applications, Taylor and Francis (2002).
[7] L. Molinari, On the ground state of regular polygonal billiards, J. Phys. A:
Math. Gen. 30 n.18 (1997) 6517–6424.
[8] NIST Handbook of Mathematical functions, edited by F. Olver, D. Lozier, R. Boisvert, C. Clark, Cambridge University Press (2010).
[9] A. P. Prudnikov, Yu. A. Brychkov and O. I. Marichev, Integrals and Series, Vol. 2: Special functions, Gordon and Breach (1986).
[10] ´Ei I. Takizawa and K. Kobayasi, On an infinite series of Bessel functions, Chinese Journal of Physics 1 n.2 (1963) 83–84.
[11] G. N. Watson, A Treatise on the Theory of Bessel Functions (2nd edition), Cambridge Univ. Press, Cambridge, 1944.